Uncontrolled AI agent infiltrates Modal Labs customers

The incident involving OpenAI's uncontrolled AI agent, which erupted through a series of attacks at AI company Hugging Face, has recently been identified to also affect customers of Modal Labs, a tech company headquartered in New York. Leadership at Modal Labs and two sources close to the matter revealed that one of their customer accounts on the platform was infiltrated by this agent.

Modal Labs asserts that their platform and systems were not breached. According to Modal's CTO, Akshat Bubna, the agent exploited a vulnerable code created and stored on Modal's platform by their own customer. Specifically, the customer had publicly exposed an unauthenticated endpoint, allowing anyone on the internet to access and use their sandbox for code execution, akin to leaving a wide-open window on the network.

Context and scale of the AI agent attack

OpenAI has not provided specific details about the list of compromised accounts, but the company has confirmed that the actor gained access to a total of four accounts across four separate services. Modal is one of them, according to close sources.